Monolith Research
Disclaimer
This was built as a joke. It is filed here because even jokes produce data, and data belongs in the record.
The premise
The hypothesis was simple enough to state at a dinner table: Trump posts on Truth Social correlate with short-term price movement in USD crosses, gold, and US equity indices. Tariff announcements, Fed commentary, China trade rhetoric — it seemed plausible that systematic keyword classification of that feed could generate a tradeable signal.
We decided to find out. This was not a serious research priority. It was a Friday afternoon idea. It went through the same pipeline as the serious ones because the pipeline does not know the difference.
Construction
The signal classified Truth Social posts by keyword density across five buckets: tariff/trade, China, Fed/rates, gold/commodities, crypto. A long or short entry on the relevant instrument was triggered when a keyword cluster exceeded a rolling threshold within a 4-hour window. Exit logic was session-based with a hard stop.
Instruments tested: EUR/USD, USD/JPY, XAU/USD, SPX via ETF proxy. The signal was applied to H4 and H1 bars with a one-bar execution lag.
Results
| Metric | Value |
|---|---|
| OOS Sharpe | −0.31 |
| Win rate | 44% |
| Max drawdown | −6.2% |
| Profit factor | 0.89 |
The signal traded noise at cost. The few genuinely predictive windows — early tariff announcements with large immediate price impact — were not repeatable enough to produce positive expectation at the aggregate level. FinBERT-style enrichment of the post text made the result marginally worse, not better.
Why it is in the file
Two reasons.
First, failed research is research. A coin-flip model, written down honestly, is cheaper than a beautiful one allowed to trade. The Trump Trader did not have edge. That is a clean, useful answer.
Second, it was genuinely funny to run. A trading strategy that depends on one person’s posting habits is not a strategy — it is a correlated bet on executive attention, which is by definition non-stationary, non-systematic, and non-repeatable. The result is appropriately negative.
Status
Closed. Not in the research queue. Archived here as a record of what a topically amusing idea looks like when it goes through the same evaluation pipeline as serious ones. The pipeline treated it with full seriousness. The seriousness was, in retrospect, itself somewhat funny.